Viscoelastic Fluid Simulation with OpenFOAM

Joan Marco Rimmek and Adrià Barja Romero

Universitat Politècnica de Catalunya. BarcelonaTECH.
(Dated: May 31, 2017)
Viscoelastic fluids do not behave like newtonian fluids and have to be explained in a different way
than classical ones. In this report we will learn a bit about the models which explain viscoelastic
fluids and we will base the project in the characterization and simulation of them acting in a
particular experiment. This experiment is the oscillation of the fluid between infinite parallel planes.
We are going to analyze and compare the obtained results with the theoretical behaviour.

INTRODUCTION well known Oldroyd-B model which represents a solution

of a Maxwellian viscoelastic fluid. The parameters in-
Complex fluids as a whole (which include viscoelastic volved in this model are: λ which is the single relaxation
fluids), do not behave as classical ones. They have an time, the rate independent viscosity ηp , and constant vis-
internal structure that makes them show non-newtonian cosity ηs . Its constitutive equations for the stress tensor
rheological properties. They need to be studied since are:
they are present in our every-day life. For instance,
blood, sauces, creams, are examples which need to be τ = τs + τp (1)
studied in order to control their flow in industry.

τs = ηs γ(1) (2)
OBJECTIVE

The objective of this work, was to learn to use the λ

complex fluid simulator OpenFOAM and the extension τp + λτp(1) − α {τp · τp } = ηp γ(1) (3)
ηp
RheoTool “an open-source toolbox based on OpenFOAM
to simulate the flow of Generalized Newtonian Fluids
(GNF) and viscoelastic fluids”. In these equations, τ is the stress tensor and α the non-
In order to accomplish it, we are going to simulate an linearity introduced in the Oldroyd-B model in order to
experiment and study some of its non-classical properties. obtain the Giesekus model. As it may seem obvious, if
With the aid of these open source software we will design α = 0 we recover the basic Oldroyd-B equations. It also
a grid and use the specific solver for the fluid under study. has to be taken into account that this solver does not only
In addition, we will extract data and compare it with the use these constitutive equations but also solves them all
theoretical results found in [1] and [2]. along with the well-known Navier-Stokes equations for
The fluid under study is a viscoelastic fluid. Un- fluids.
like classical Newtonian fluids, which present a constant We are not going to analyze these equations but just
viscosity, complex fluids (and in particular, viscoelastic present them. In the simulation, these equations are effi-
ones) exhibit rate dependant shear viscosity, which will ciently implemented by the solver included in RheoTool,
cause different behaviours at different time-scales. This so we will not need to worry about them.
change in behaviour can be explained through the Deb-
orah number, but we are not going to enter in detail on
the theory. More information about complex fluids can
SETUP
be found in [3].
The experiment we are going to analyze is the response
of the fluid to an harmonic movement between two par- A. General aspects
allel infinite plates. Since the resonant frequencies can
be theoretically found [1], we are going to perform the It would be great to simulate the behaviour of a fluid
same simulation for two resonant frequencies and one in oscillating in a cylindrical pipe in order to compare it
between, to see the response for different forcings of the with the empirical results (apart from the theoretical
system. ones). However, as we mentioned above, we will simu-
late the behaviour of a viscoelastic fluid subjected to a
sinusoidal movement between two infinite plates as a first
MODEL approximation to the cylindrical case.
The different steps we followed to perform the simula-
The theoretical model we are going to use in this sim- tions in OpenFOAM along with the RheoTool library are
ulation is known as Giesekus model. It derives from the explained in the next sections.
 2

B. Geometry very similar case studied in [1]. Those can be calculated

as:
First of all we need a geometry for the simulation to ηp π 2 (1 + 2n)2 ηp π 2 (1 + 2n)2 λ − 1
take place in. The geometry needs to be translated into ω0 = 2
(1 + ) 2 (6)
2a ρ a2 ρ
a grid in an understandable way. In order to do so we
We will perform the simulation in two resonant cases
generate a mesh using the BlockMesh utility with the
and one in between:
following configuration:
We generate a square based prism of dimensions
(0.3, 0.05, 0.05)m. The x axis is the one in which the ω01 = 11.250704Hz ω03 = 33.760313Hz
pistons will move. For this prism to become what we
want (two infinite planes) we set a symmetry condition ω02 = 22Hz
in the z axis. We will call a = 0.0025m the half-width of
the distance between the infinite plates.
Furthermore, this geometry has to be discretized, as RESULTS
the x and y axis are the most important in our case, we
set the number of divisions to be (60, 60, 16). A figure of E. General aspects
the resulting grid can be found in Annex II.
In this section we are going to analyze the results ob-
tained from the simulation and compare them to theoret-
C. Boundary conditions and initial conditions ical external results. The main results are obtained for a
linear viscoelastic fluid, so for α = 0.
The best way we found to simulate an oscillating piston What we are studying in this section is the profile of
was to impose a sinusoidal movement at the inlet and velocity parallel to the infinite plates (namely the direc-
the outlet for the velocity of the fluid. To do so, we will tion of the oscillation) along the perpendicular direction
impose the following function in time at x = 0 (the fluid between the plates (y axis). This will allow us to eas-
first entering the grid) and at x = 0.3 (first leaving the ily compare the changes in the simulation. The point
grid). In Annex II we can see the flow of the fluid at the of study will be the center of the grid, x = 0.15m and
initial time. x = 0.6m for the long case.
We are going to make the same measurements for dif-
U (t) = x0 ω0 cos(ω0 t) (4) ferent cases. On one hand we are going to study if the
simulation reaches laminar flow at the center for a certain
Which is trivially derived from: longitude, that is to say, we will simulate two different x
lengths and see how this affects the results.
x(t) = x0 sin(ω0 t) (5)
On the other hand we are going to study the effects of
Where x0 is the amplitude of the oscillation and ω0 the discretization. First we are going to simulate with a
is its frequency. Observe that since cos(0) = 1 the fluid thick grid and then with a thinner one.
starts in movement at the boundaries. As a last step of the whole project we are going to
The initial condition for the fluid will be of zero veloc- analyze the case with α = 0.85.
ity except at the boundaries. All the numerical results are obtained after some pe-
riods of the oscillatory forcing, when the system has
reached a steady state i. e. the velocity profile does
D. Simulation parameters not change after a integer number of periods.

The parameters used for the viscoelastic fluid are the F. Theoretical results
ones which will allow us to compare with the empirical
results for a specific viscoelastic fluid:
The theoretical result with which we will compare are
obtained by [2]:
Kg
λ = 1.9s ηp = 64 m·s kL cosh(ky)
v(y, ω) = dv (1 − ) (7)
kL − tanh(kL) cosh(kL)
ρ = 1050 Kg
m3
Kg
ηs = 0 m·s
v(y, t) = Re{v(y, ω) exp(iωt)} (8)
We will first simulate the linear case with α = 0 and p
then α = 0.85. with dv = z0 ω and k = (iωρ/η)(1 + iλω).
The amplitude we will use in our simulation is x0 = In figure [1] we can see the plot obtained with these
0.001m. equations for different frequencies.
There is no specific way in the references to calculate The x axis represents the position between the plane
the resonant frequencies but we can use the ones for a and the y axis the velocity.
 3

Figure 1: Theoretical U profile for different frequencies

for different times in the same period. From left to
right: ω01 , ω03 and ω02
Figure 3: Simulation for ω03 (2nd resonant frequency) for
different times in the same period
G. Simulation results

The results obtained for the simulation in the same

conditions has figure [1] are:

Figure 4: Simulation for ω02 (non resonant frequency)

for different times in the same period

Figure 2: Simulation for ω01 (1st resonant frequency) for

different times in the same period

We can see how ω01 qualitatively behaves as the the-

oretical results predict whilst ω02 and ω03 do not match
the predictions. In order to explain this we are going to
analyze some variables of the simulation.

H. Longitude analysis

We are going to simulate again the second resonant

frequency to observe the differences when longitude is
changed. We should expect that the finiteness of the sim-
ulation will affect it and thus, elongating the grid should Figure 5: Longer grid simulation for ω03 (2nd resonant
improve the results. However as we can see comparing frequency) for different times in the same period
figures [3] and [5] we can see a change in amplitude and
also the nodes at the centre seem to displace less from
U = 0, but the results do not behave as the theoretical The difference in the amount of lines is there because
ones. in the moment that we performed the simulation we just
 4

graphed a different amount of lines, it has nothing to do K. Convergence

with the simulation in itself.
In every one of the cases aforementioned, we checked
that the system reached steady state, comparing how the
I. Fineness analysis velocity profile evolved at the maximum velocity of oscil-
lation in each period (in a particular direction). Obtain-
ing a graph like [8].
It is well known that in most cases increasing the fine-
ness does not affect the results. However, in this simu-
lation the nodes of the graph seem to displace less from
U = 0 as we increase the number of cells in the grid. But
the results still do not resemble the predictions.

Figure 8: Convergence plot

Normally, we reached steady state after 5 or 6 periods.

Figure 6: Thin grid simulation for ω03

CONCLUSIONS

First of all we should say that making such a simulator

J. Non-linear case work was not as easy as it may seem. Installing all the
dependencies, understanding how to input the model, the
Just to see how a more complex model would behave in parameters and the grid, and then extracting the infor-
such simulations, we simulated the non-linear case with mation that we can see in the plots was difficult given the
α = 0.85 obtaining similar results (figure [7]). limited documentation available of the software. How-
ever, we managed to do it and obtain the results shown.
The simulation results, though coherent between them,
do not coincide with the theoretical results at all. We
have thought of different reasons for why this could hap-
pen and reached some conclusions.
Both increasing the separation of the pistons and the
thinness of the grid alters a little bit the amplitude and
makes the profiles pass nearer U = 0 at the middle point
between the parallel plates as in the theoretical results.
However this does not seem to tend to the desired result.
Maybe increasing the thinness of the grid a lot more could
end up converging to the theoretical result but it does not
look promising.
There are two main reasons that we think may explain
the obtained results. On one hand, the simulator might
not be doing what we wanted at all. A thorougher study
Figure 7: Simulation for ω03 (2nd resonant frequency) and work with the software would have been great and
and α = 0.85 for different times in the same period we could have a better performance of the simulation,
though it was not possible with limited time. On the
other hand we may have wronged by approximating the
 5

boundary conditions of the pistons with the function in about the scientific method and the trial and error which
velocities. characterizes it. We would also like to thank L. Ramı́rez-
Finally we would like to that say we have learned a lot Piscina for conducting such an interesting project.

REFERENCES January 2013.

[2] L. Ramı́rez-Piscina via private communication.
[1] Laura Casanellas Vilageliu. Oscillatory pipe flow of worm- [3] Alexander Morozov and Saverio E. Spagnolie. Introduction
like micellar solutions, PhD Thesis, Physics Dept., UB. to Complex Fluids (book)
 ANNEX I
Adrià Barja Romero and Joan Marco Rimmek
May 2017

Introduction
In this annex, the most important code files are presented. They define the
properties of the fluid, the configuration of the simulation, the geomentry and
mesh used and the initial conditions.

Properties of the fluid

 constant/

In this folder you can find the dictionary constitutiveProperties.

In this file, the properties of the fluid are defined. In the example shown
bellow the properties where defined by: ηP = 64 m·s kg
, ηS = 0, ρ = 1050 m
kg
3 and

λ = 1.9s.

1 /∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗− C++ −∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗\

2 | ========= | |
3 | \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
4 | \\ / O peration | Version : 4.0 |
5 | \\ / A nd | Web: www.OpenFOAM. org |
6 | \\/ M anipulation | |
7 \∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗/
8 FoamFile
9 {
10 version 2.0;
11 format ascii ;
12 class dictionary ;
13 object constitutiveProperties ;
14 }
15 // ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ //
16
17 parameters
18 {
19 type Oldroyd−BLog;
20
21 rho rho [ 1 −3 0 0 0 0 0 ] 1050;
22 etaS etaS [ 1 −1 −1 0 0 0 0] 0;
23 etaP etaP [ 1 −1 −1 0 0 0 0 ] 64;
24 lambda lambda [ 0 0 1 0 0 0 0] 1.9;
25 uTauCoupling true ;
26 }
27
28 passiveScalarProperties

1
29 {
30 solvePassiveScalar off ;
31 D D [ 0 2 −1 0 0 0 0 ] 1e−9;
32 }
33 // ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗ //

The numbers between brackets define the units of each constant, they cor-
respond to [kg m s K mol A cd].

Initial conditions
 0/

In this folder there are four files: p, tau, theta and U.

We will show the file U since it is the most interesting in our case. The syntax
of the others is very similar in any case.
Bellow is the initial condition for an oscillation of amplitude 0.001m and
frequency of 33.76...rad/s.
1 /∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗− C++ −∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗\
2 | ========= | |
3 | \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
4 | \\ / O peration | Version : 4.0 |
5 | \\ / A nd | Web: www.OpenFOAM. org |
6 | \\/ M anipulation | |
7 \∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗/
8 FoamFile
9 {
10 version 2.0;
11 format ascii ;
12 class volVectorField ;
13 object U;
14 }
15 // ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ //
16
17 dimensions [ 0 1 −1 0 0 0 0 ] ;
18
19 internalFiel d uniform (0 0 0) ;
20
21 boundaryField
22 {
23 inlet
24 {
25 type codedFixedValue ;
26 value uniform (0 0 0) ;
27 redirectType smoothU ;
28 code
29 //Cosinusoidal
30 #{
31 const s c a l a r& t = this−>db( ) . time ( ) . timeOutputValue ( ) ;
32 vector Uav(1 , 0 , 0) ;
33 vector dirN (1 , 0 , 0) ;
34 Uav = ((0.001∗33.760313226149∗Foam : : cos (33.760313226149∗ t ) ) ∗ dirN ) ;
35 operator == (Uav) ;
36 #};
37 }
38
39 walls
40 {
41 type fixedValue ;

2
42 value uniform (0 0 0) ;
43 }
44
45 outlet
46 {
47 type codedFixedValue ;
48 value uniform (0 0 0) ;
49 redirectType smoothU ;
50 code
51 //Cosinusoidal
52 #{
53 const s c a l a r& t = this−>db( ) . time ( ) . timeOutputValue ( ) ;
54 vector Uav(1 , 0 , 0) ;
55 vector dirN (1 , 0 , 0) ;
56 Uav = ((0.001∗33.760313226149∗Foam : : cos (33.760313226149∗ t ) ) ∗ dirN ) ;
57 operator == (Uav) ;
58 #};
59 }
60
61 frontAndBack
62 {
63 type symmetry ;
64 }
65 }
66 // ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗ //

Mesh properties
 system/

In this folder we can find the file blockMeshDict which dictates the geometry
in which the simulation will take place.
1 /∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗− C++ −∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗\
2 | ========= | |
3 | \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
4 | \\ / O peration | Version : 4.0 |
5 | \\ / A nd | Web: www.OpenFOAM. org |
6 | \\/ M anipulation | |
7 \∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗/
8 FoamFile
9 {
10 version 2.0;
11 format ascii ;
12 class dictionary ;
13 object blockMeshDict ;
14 }
15 // ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ //
16
17 convertToMeters 0 . 1 ;
18
19 vertices
20 (
21 (0 −0.25 −0.25)
22 (3 −0.25 −0.25)
23 (3 0.25 −0.25)
24 (0 0.25 −0.25)
25 (0 −0.25 0.25)
26 (3 −0.25 0.25)
27 (3 0.25 0.25)
28 (0 0.25 0.25)
29 );
30

3
31 blocks
32 (
33 hex (0 1 2 3 4 5 6 7) (50 60 12) simpleGrading (1 1 1) //0
34 );
35
36 edges
37 (
38 );
39
40 boundary
41 (
42 inlet
43 {
44 type patch ;
45 faces
46 (
47 (0 4 7 3)
48 );
49 }
50
51 outlet
52 {
53 type patch ;
54 faces
55 (
56 (1 2 6 5)
57 );
58 }
59
60 walls
61 {
62 type wall ;
63 faces
64 (
65 (2 3 7 6)
66 (0 1 5 4)
67 );
68 }
69
70 frontAndBack
71 {
72 type symmetry ;
73 faces
74 (
75 (4 5 6 7)
76 (1 0 3 2)
77 );
78 }
79 );
80
81 mergePatchPairs
82 (
83 );
84 // ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗ //

The option convertToMeters 0.1 multiplies each dimension by 0.1. There-

fore, our parallel planes have a separation of 0.025m and the distance between
the two pistons is of 0.3m.

4
 Simulation parameters
 system/

Also, in folder system we can find the file controlDict which dictates the
parameters of the the simulation in itself. Here you can configure the initial and
final times, the time-step and personalized functions in order to retrieve data
from the raw files.
1 /∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗− C++ −∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗\
2 | ========= | |
3 | \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
4 | \\ / O peration | Version : 4.0 |
5 | \\ / A nd | Web: www.OpenFOAM. org |
6 | \\/ M anipulation | |
7 \∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗/
8 FoamFile
9 {
10 version 2.0;
11 format ascii ;
12 class dictionary ;
13 object controlDict ;
14 }
15 // ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ //
16
17 application rheoFoam ;
18
19 startFrom startTime ;
20 startTime 0;
21
22 stopAt endTime ;
23 endTime 0.95;
24
25 deltaT 0.0002;
26
27 writeControl timeStep ;
28 writeInterval 1;
29 purgeWrite 0;
30 writeFormat ascii ;
31 writePrecision 12;
32 writeCompression compressed ;
33 timeFormat general ;
34 timePrecision 10;
35 graphFormat raw ;
36 runTimeModifiable no ;
37 adjustTimeStep o f f ;
38 maxCo 0.1;
39 maxDeltaT 0.01;
40
41 functions
42 {
43 #includeFunc singleGraph
44 }
45 // ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗ //

In this case we start the simulation at time 0s and end it at time 0.95s using
a time-step of 0.0002s. Moreover, the function singleGraph can be found in
another file in the same folder, and is the one that retrieves the data of the
velocity profiles:

5
 1 /∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗− C++ −∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗\
2 ========= |
3 \\ / F ield | OpenFOAM: The Open Source CFD Toolbox
4 \\ / O peration |
5 \\ / A nd | Web: www.OpenFOAM. org
6 \\/ M anipulation |
7 −−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
8 Description
9 Writes graph data f o r s p e c i f i e d f i e l d s along a l i n e , s p e c i f i e d by s t a r t
10 and end points .
11 \∗−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−∗/
12
13 start (0.15 −0.025 0) ;
14 end (0.15 0.025 0) ;
15 fields (U) ;
16
17 // Sampling and I /O s e t t i n g s
18 #includeEtc ” caseDicts/postProcessing/graphs/sampleDict . cfg ”
19
20 // Override s e t t i n g s here , e . g .
21 // setConfig { type midPoint ; }
22
23 // Must be l a s t entry
24 #includeEtc ” caseDicts/postProcessing/graphs/graph . cfg ”
25 // ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗ //

Although there are many more files in each OpenFOAM case, those where
the most important and the only ones we had to modify to make one simulation
or another.

6
 ANNEX II
Adrià Barja Romero and Joan Marco Rimmek
May 2017

Introduction
In this annex images and plots obtained from the simulation are present.

Grid
The grid we used in our simulations, which is 0.3m long and has 60 divisions in
the x axis, is 0.05m long and has 60 divisions in the y axis and is 0.05m long
and has 16 divisions in the z axis.

Figure 1: Grid used in the simulations

1
Post-processed simulations

Figure 2: Image of the post-processed simulation at the first moments of the

simulation

Figure 3: Image of the post-processed simulation at a change of direction in the

velocity